I got a point cloud representing the early medieval sites in Brandenburg (about 7000). They cluster roughly like displayed below.
There are some articles (e.g. Nakoinz 2015) about how to deduce networks from clusters like this, by generating cluster centroids and using a Delaunay-triangulation on those. However, they usually don't go into too much detail. Has anyone worked with something like this before or can give me hints on literature or methodology.
What I would like to get in the end is an approximation of the most likely spatial network, which I would like to compare to the cost surface of the area. My guess would be that network connections would be stronger along paths with lower costs.
Kommentar
Lukas Goldmann sagt: Autor
After the comments I got during the last meeting, I did some point clustering using DBScan in QGIS. This worked quite good, as displayed below.
(From left to right: Points clustered by DBScan (grey = no cluster/noise); Clusters converted into minimal bounding geometries;Centroids of those bounding geometries)
The clusters more or less fit to the groups we already know from written sources, but are of course biased by different states of research. However, this is the nature of our data. Also, one has to choose the size of a cluster "kernel", here I chose 5 km, based on the assumption that this is about the 1h walking distance a human can cover. This radius has often been used therefore in site catchment analyses and other archaeological spatial analyses.
This work flow produces a model I can work/live with. It has it's problems of course, but which model doesn't.